The network was trained using public images generated by users of the Imgflip Meme Generator for the top 48 most popular Meme Templates. Using someones name or other short text as a prefix works best. Computational photography is already present in all modern smartphones to some degree, and it's a question of time before algos like this are fully integrated as well. The model uses character-level prediction, so you can specify prefix text of one or more characters to influence the text generated. The algorithms run on raw image data and adds AI magic to figure out what footage "should have" looked like without the distinct video noise generated by imaging sensors.Īt the moment this is research rather than a commercially available product, but as a photography and AI nerd, I'm wildly excited by these developments the lines are blurring between photography and computer graphics, and I'm here for it.
I can write a million words about how awesome this is, but I can do better here's a 1-minute-51-second video, which, at 30 frames per second and "a picture tells a thousand words," is at least 1.5 million words worth of magic: But then Google Research came along, releasing an open source project it calls MultiNerf, and I get the sense that we're at the precipice of everything changing.
We may be getting close to what is possible to achieve with hardware heat and physics are working against us making even better camera sensors. Sure, some cameras can shoot haunting video lit only by the light of the moon, but for stills - and especially stills shot on a smartphone - digital noise continues to be a scourge. If you have great lighting, a good photographer can take decent photos even with the crappiest camera imaginable.
0 kommentar(er)
